The possible impact of weather on crimes in Amsterdam

Show code cell source Hide code cell source

import pandas as pd
from pandas import Timedelta
import numpy as np
import plotly.express as px
import seaborn as sns
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
from IPython.display import display, Markdown
import plotly.io as pio
pio.renderers.default = "notebook"
import plotly.graph_objects as go
from scipy import stats
df = pd.read_csv("merged_weather_misdrijven_monthly_v3.csv")

People have long studied both weather and human behavior. A key question at their intersection is whether weather influences crime.

Some researchers think so: for instance, Anderson (2001) found that high temperatures may increase aggression. In this project, we examine whether weather factors like heat or rain are linked to crime rates in Amsterdam.

We use daily weather data from Schiphol Airport (KNMI, 2025) and monthly crime data from Statistics Netherlands (CBS, 2025). This allows us to explore both seasonal patterns in serious crimes and specific links, such as between warmer days and bike theft.

Using data visualizations, we assess whether crime is weather-dependent or largely unaffected by it.

Perspective 1: Pleasant weather increases the risk of property and substance related crimes

Pleasant weather often brings people outside, increasing public activity in parks, streets, and recreational areas. This rise in outdoor presence may also create more opportunities for crimes such as bicycle theft, pickpocketing, or alcohol-related incidents. In this section, we examine how sunshine and higher temperatures might correlate with these types of crime.

Argument 1.1 Higher temperatures lead to more bicycle theft

In Amsterdam, cycling is the main form of transport. Better weather conditions encourage more outside activity such as events or going to the beach, which means more transport and bikes being left outside. This should be the optimal climate for bicycle theft and supports the idea that weather has an influence on crime. To look into this relationship between theft and tropical weather, we inspect temperature and the rates of bicycle theft in the graph below.

Show code cell source Hide code cell source

df.columns = df.columns.str.strip()
df['TX'] = df['TX'] / 10  # TX is in tienden van °C
df['TG'] = df['TG'] / 10  # TG is in tienden van °C
df['SQ'] = df['SQ'] / 10  # TG is in tienden van °C
display(Markdown("### Bicycle Theft vs. Maximum Temperature"))
display(Markdown("_Slight increase in thefts when it's warmer._"))

# 📈 Plot maken
plt.figure(figsize=(8, 5))
sns.regplot(
    x='TX',
    y='1.2.3 Diefstal van brom-, snor-, fietsen',
    data=df,
    scatter_kws={'alpha': 0.5}
)
plt.xlabel("Maximum Temperature (°C)")
plt.ylabel("Number of Bicycle Thefts")
plt.grid(True)
plt.tight_layout()
caption = (
    "Figure: Each dot represents monthly bicycle thefts and maximum temperature. "
    "The blue line shows a moderate positive trend between temperature and thefts."
)
plt.figtext(0.5, -0.1, caption, wrap=True, horizontalalignment='center', fontsize=10)

plt.show()

Bicycle Theft vs. Maximum Temperature

Slight increase in thefts when it’s warmer.

This scatterplot shows the relationship between monthly maximum temperature (°C) on the x-axis and reported bicycle thefts on the y-axis. Each point represents one month. The blue regression line shows a slight upward trend: bike thefts increase as temperatures rise. Pearson’s correlation is 0.502, considered moderate. To put it intuitively, each degree increase corresponds to an average of 13.23 more bicycles stolen. This supports the first argument that warmer weather relates to more bike theft.

Argument 1.2 Substance related incidents on boats increase with sunshine

As days get warmer and sunnier, more people take to the water. Amsterdam’s canals are especially busy when the sun shines bright. This creates more opportunities for things to go sideways, suggesting weather does affect crime. We examine this by comparing sunshine hours with water-related incidents under influence.

Show code cell source Hide code cell source

import matplotlib.dates as mdates
from pandas import Timedelta
from matplotlib.ticker import FuncFormatter

# Titel en korte take-away boven de plot
display(Markdown("### Weather vs. Water-related under influence reports"))
display(Markdown("_More sunshine coincides with more 'under influence on boat' incidents._"))

df['year_month'] = pd.to_datetime(df['year_month'])

fig, ax = plt.subplots(figsize=(12,6))

# Lijnen plotten
ax.plot(df['year_month'], df['SQ'], label='Sunshine duration (Hours)', color='blue')
ax.plot(df['year_month'], df['3.4.2 Onder invloed (water)'], label='Under influence on boat (#Incidents)', color='green')

# As-labels en titel
ax.set_xlabel('Year', fontsize=16)
ax.set_ylabel('Value', fontsize=16)


# Zet x-as ticks alleen per jaar (niet per maand)
years = df['year_month'].dt.year.unique()
year_starts = [pd.Timestamp(f'{year}-01-01') for year in years]
ax.set_xticks(year_starts)

# Labels leeg maken, want we tonen jaartallen onder de as in blokken
ax.set_xticklabels(['' for _ in year_starts])

# Maak blokken per jaar (lichte achtergrond)
for year in years:
    start = pd.Timestamp(f'{year}-01-01')
    end = pd.Timestamp(f'{year}-12-31')
    ax.axvspan(start, end, color='lightgrey', alpha=0.2)

# Bepaal min en max van data
min_y = min(df['SQ'].min(), df['3.4.2 Onder invloed (water)'].min())
max_y = max(df['SQ'].max(), df['3.4.2 Onder invloed (water)'].max())
margin = 0.1 * (max_y - min_y)
# Maak y-limiet iets lager zodat haakjes zichtbaar zijn, minimaal -0.2 bijvoorbeeld
lower_lim = min(-0.2, min_y - margin)
upper_lim = max_y + margin
ax.set_ylim(lower_lim, upper_lim)

# Haak-achtige lijnen en jaartallabels onder de x-as
ylim = ax.get_ylim()
# Y_base iets lager zodat haakjes in zicht blijven
y_base = ylim[0] - 0.03*(ylim[1]-ylim[0])    # Iets dieper dan vorige -0.05, zodat haakjes echt zichtbaar zijn
y_hook_top = y_base + 0.02*(ylim[1]-ylim[0])

for year in years:
    start = pd.Timestamp(f'{year}-01-01')
    end = pd.Timestamp(f'{year}-12-31')

    # horizontale lijn (onder het jaarblok)
    ax.hlines(y=y_base, xmin=start, xmax=end, colors='black', linewidth=1.5)

    # verticale lijntjes aan begin en eind als haakjes
    ax.vlines(x=start, ymin=y_base, ymax=y_hook_top, colors='black', linewidth=1.5)
    ax.vlines(x=end, ymin=y_base, ymax=y_hook_top, colors='black', linewidth=1.5)

    # jaarlabel gecentreerd tussen de haakjes, iets onder de lijn
    mid = pd.Timestamp(f'{year}-07-01')
    ax.text(mid, y_base - 0.02*(ylim[1]-ylim[0]), str(year), ha='center', va='top', fontsize=12)

# Rasterlijnen alleen horizontaal (geen verticale maandlijnen)
ax.grid(axis='y')

# Custom formatter: geen negatieve y-as labels tonen
def positive_only_formatter(x, pos):
    return f'{int(x)}' if x >= 0 else ''

ax.yaxis.set_major_formatter(FuncFormatter(positive_only_formatter))

# Legend en layout
ax.legend()


fig.tight_layout()

# X-as limieten iets uitbreiden
start_lim = df['year_month'].min() - Timedelta(days=20)
end_lim = df['year_month'].max() + Timedelta(days=300)
ax.set_xlim(start_lim, end_lim)
caption = (
    "Figure: Monthly sunshine duration (blue line) and reports of being under influence on boats (green line). "
    "Data points show trends over time from year to year.")
plt.text(0.5, -0.1, caption, ha='center', va='top', fontsize=10, transform=ax.transAxes)

plt.show()

Weather vs. Water-related under influence reports

More sunshine coincides with more ‘under influence on boat’ incidents.

This multivariate line chart compares sunshine duration (hours) and the number of water-related substance use incidents (“under influence on water”) per month. The horizontal axis shows months; the vertical axis shows the values. The blue line indicates that more sunshine months coincide with more incidents. Pearson’s correlation is 0.457, suggesting a moderate relationship. A peak in summer 2022 may reflect increased outdoor activity post-COVID lockdowns, suggesting weather influences such behavior.

It is clear that weather can be a predictor of both substance related incidents on water and bicycle theft, however, those are quite niche categories of crime. To get a better understanding of the general relationship between pleasant weather and crime, we use a heatmap for visualisation that includes more categories.

Show code cell source Hide code cell source

weer_vars = ['TG', 'RH', 'VVN']
misdaad_vars = [
    '1.1.1 Diefstal/inbraak woning',
    '1.3.1 Ongevallen (weg)',
    '3.4.2 Onder invloed (water)',
    '1.4.2 Moord, doodslag',
    '2.5.2 Winkeldiefstal',
    '1.2.4 Zakkenrollerij',
    '1.2.3 Diefstal van brom-, snor-, fietsen'
]

# Subset dataframe
df_subset = df[weer_vars + misdaad_vars]

# Correlatiematrix
corr_matrix = df_subset.corr()
corr_submatrix = corr_matrix.loc[weer_vars, misdaad_vars]

# Vertalingen
crime_translation = {
    '1.1.1 Diefstal/inbraak woning': 'Burglary (home)',
    '1.3.1 Ongevallen (weg)': 'Traffic Accidents',
    '3.4.2 Onder invloed (water)': 'Under Influence (water)',
    '1.4.2 Moord, doodslag': 'Murder, manslaughter',
    '2.5.2 Winkeldiefstal': 'Shoplifting',
    '1.2.4 Zakkenrollerij': 'Pickpocketing',
    '1.2.3 Diefstal van brom-, snor-, fietsen': 'Bicycle Theft'
}

weather_translation = {
    'TG': 'Avg. Temperature\n(°C)',
    'RH': 'Precipitation\n(mm)',
    'VVN': 'Min Visibility\n(hm)'
}

corr_submatrix.columns = corr_submatrix.columns.map(crime_translation)
corr_submatrix.index = corr_submatrix.index.map(weather_translation)

# Display titel en caption
from IPython.display import display, Markdown

plt.figure(figsize=(12, 6))
ax = sns.heatmap(
    corr_submatrix,
    annot=True,
    cmap='coolwarm',
    center=0,
    vmin=-0.6,
    vmax=0.6
)

plt.title('Correlation between Weather Variables and Crime Categories', fontsize=16)
plt.xlabel('Crime Categories', fontsize=14)
plt.ylabel('Weather Variables', fontsize=14)
plt.xticks(rotation=45, ha='right')
plt.yticks(rotation=0)
plt.grid(False)

# Caption onderaan de plot
caption = (
    "Heatmap shows how each weather variable correlates with different crime types. "
    "Only a few crimes show meaningful correlations with weather."
)
plt.gcf().text(0.5, -0.1, caption, ha='center', fontsize=10)

plt.tight_layout()
plt.show()

The vertical axis represents weather conditions: minimum visibility (hm) (for example because of fog), precipitation (mm) and average temperature (C°). The horizontal axis represents crime types. This heat map shows that out of those categories only ‘under influence (water)’ and ‘bicycle theft’ have substantial correlations with any weather conditions at all. This means that the correlation between crime and warm weather is limited to those categories. Confirming our arguments, and our perspective.

Perspective 2: Serious crimes are not affected by bad weather

An alternative perspective shows that, although specific categories of crime may be affected by warm weather conditions, serious crimes show no correlation with weather at all. In this section, we explore how rainfall might relate to burglary, and whether certain types of crime show seasonal patterns, regardless of temperature.

Argument 2.1: Increased rainfall has no correlation to the amount of burglaries

Although one might expect rain to affect burglary rates, research from Beijing (Peng et al., 2011) found no significant link between precipitation and residential burglaries. This suggests that increased rainfall may not directly influence burglary incidents. Our analysis aims to explore whether this holds true in our data.

Show code cell source Hide code cell source

# Titel en korte take-away boven de plot
display(Markdown("### Rainfall vs. Home Burglaries"))
display(Markdown("_There appears to be no clear correlation between monthly rainfall and burglary rates._"))

plt.figure(figsize=(10,6))

sns.regplot(
    data=df,
    x='RH',  # gemiddelde neerslag in mm
    y='1.1.1 Diefstal/inbraak woning',
    scatter_kws={'s':50, 'alpha':0.6},
    line_kws={'color':'red'}
)

plt.xlabel('Average Monthly Rainfall (mm)', fontsize=14)
plt.ylabel('Total Burglary Incidents per Month', fontsize=14)
plt.grid(True)

# Caption onderaan, gecentreerd en iets kleiner
caption = (
    "Scatterplot shows monthly data points with average rainfall on the x-axis "
    "and total burglary incidents on the y-axis. The red line represents the regression trend."
)
plt.gcf().text(0.5, -0.1, caption, ha='center', fontsize=10)

plt.tight_layout()
plt.show()

# Bereken correlatie en helling
correlation, p_value = stats.pearsonr(df['RH'], df['1.1.1 Diefstal/inbraak woning'])
slope, intercept, r_value, p_value_reg, std_err = stats.linregress(df['RH'], df['1.1.1 Diefstal/inbraak woning'])

Rainfall vs. Home Burglaries

There appears to be no clear correlation between monthly rainfall and burglary rates.

This scatterplot shows average monthly rainfall (x-axis) versus total burglary incidents per month (y-axis). The red regression line indicates the trend, but with a Pearson correlation of just 0.077, there is almost no linear relationship. This suggests that increased rainfall has little to no effect on burglary rates, confirming the study from Peng et al. and our argument.

Argument 2.2: Serious crimes are not weather-dependent

As for more serious crime, generally speaking, there is no obvious reason why it should depend on seasonal weather. Murder, abuse and cybercrime should not be associated with seasonal patterns. Warm days may lead to more activity and opportunity to commit these crimes, while cold days in turn may enable more of a melancholic mental state that motivates them more. Cybercrime sounds like an indoor activity, creating the expectation of more cybercrime in the winter, which we will analyse now.

Show code cell source Hide code cell source

# Map each month number to a season
season_map = {
    1: 'Winter', 2: 'Winter', 12: 'Winter',
    3: 'Spring', 4: 'Spring', 5: 'Spring',
    6: 'Summer', 7: 'Summer', 8: 'Summer',
    9: 'Autumn', 10: 'Autumn', 11: 'Autumn'
}

# Extract numeric month from 'year_month' and map to season
df['month'] = pd.to_datetime(df['year_month']).dt.month
df['Season'] = df['month'].map(season_map)

# Serious crime columns and their display names
display(Markdown("### Seasonal Distribution of Serious Crimes"))
display(Markdown("_Serious crimes like murder, abuse, and cybercrime show relatively steady distribution across seasons._"))

serious_crimes = {
    '1.4.2 Moord, doodslag': 'Murder, manslaughter',
    '1.4.5 Mishandeling': 'Abuse',
    '3.7.4 Cybercrime': 'Cybercrime'
}

fig, axes = plt.subplots(1, 3, figsize=(24, 9))  # Nog groter

for ax, (col, label) in zip(axes, serious_crimes.items()):
    season_totals = df.groupby('Season')[col].sum().reindex(['Winter', 'Spring', 'Summer', 'Autumn'])
    wedges, texts, autotexts = ax.pie(
        season_totals,
        labels=season_totals.index,
        autopct='%1.1f%%',
        startangle=90,
        colors=sns.color_palette("pastel")[0:4],
        textprops={'fontsize': 18}  # Grotere labels en percentages
    )
    ax.set_title(f'Seasonal Distribution of {label}', fontsize=22, fontweight='bold')

# Caption groter en iets lager
caption = (
    "Each pie chart shows the proportion of incidents per season for three serious crimes. "
    "Values are total counts summed over all years. "
    "Colors represent Winter, Spring, Summer, and Autumn respectively."
)
fig.text(0.5, -0.09, caption, ha='center', fontsize=18)

plt.tight_layout(rect=[0, 0.12, 1, 0.95])  # Meer ruimte onderin voor caption
plt.show()

Seasonal Distribution of Serious Crimes

Serious crimes like murder, abuse, and cybercrime show relatively steady distribution across seasons.

These pie charts show the seasonal distribution of monthly incidents for murder, abuse, and cybercrime. If these crimes were season-dependent, large differences would be visible, especially between summer and winter. However, the distributions are fairly even across all seasons. Winter appears slightly lower due to February having fewer days. Cybercrime might occur a bit more in winter, but overall, no substantial seasonal pattern is visible. This chart supports the claim that bad weather has little influence on serious crimes.

Conclusion

As shown earlier, certain crime types, like bike theft or substance-related incidents on boats, show notable correlations with weather. Warmer and brighter days can influence behavior in ways relevant to crime prediction. In contrast, serious crimes show little to no weather dependency. To highlight this, we compare how various crimes relate to monthly maximum temperature alongside ‘under influence on water’ incidents.

Show code cell source Hide code cell source

import plotly.graph_objects as go
from IPython.display import display, Markdown

crime_columns = {
    'Burglary (home)': '1.1.1 Diefstal/inbraak woning',
    'Traffic Accidents': '1.3.1 Ongevallen (weg)',
    'Under Influence (water)': '3.4.2 Onder invloed (water)',
    'Murder, manslaughter': '1.4.2 Moord, doodslag',
    'Shoplifting': '2.5.2 Winkeldiefstal',
    'Pickpocketing': '1.2.4 Zakkenrollerij',
    'Bicycle Theft': '1.2.3 Diefstal van brom-, snor-, fietsen'    
}

def frange(start, stop, step):
    while start <= stop:
        yield round(start, 1)
        start += step

def create_slider_plot(df):
    temps = [round(t, 1) for t in list(frange(5.0, 26.5, 0.5))]
    fig = go.Figure()

    # Voeg alle frames toe, één per temperatuur
    frames = []
    for temp in temps:
        lower, upper = temp - 0.5, temp + 0.5
        filtered = df[(df['TX'] >= lower) & (df['TX'] < upper)]
        y = [
            (filtered[col].sum() / df[col].sum()) * 100 if df[col].sum() > 0 else 0
            for col in crime_columns.values()
        ]
        frames.append(go.Frame(
            data=[go.Bar(x=list(crime_columns.keys()), y=y)],
            name=f"{temp}"
        ))

    # Voeg eerste data toe als initiele trace
    fig.add_trace(frames[0].data[0])

    # Zet layout, slider en frames
    fig.update_layout(
        yaxis=dict(range=[0, 25], title="Pct of Total Incidents"),
        xaxis_title="Crime Type",
        width=800,
        height=600,
        margin=dict(b=130),  # Extra ruimte onderaan voor slider
        updatemenus=[dict(
            type="buttons",
            showactive=False,
            buttons=[dict(label="Play", method="animate",
                          args=[None, {
                              "frame": {"duration": 300, "redraw": True},
                              "fromcurrent": True,
                              "transition": {"duration": 0}
                          }])]
        )],
        sliders=[dict(
            active=temps.index(5.0),
            currentvalue={"prefix": "Temp: "},
            pad={"t": 80},  # Verplaatst slider omlaag
            steps=[dict(label=f"{t}°C", method="animate", args=[[str(t)], {
                "frame": {"duration": 0},
                "mode": "immediate"
            }]) for t in temps]
        )]
    )

    fig.frames = frames
    return fig

# Titel en korte conclusie boven de interactieve plot
display(Markdown("### Crime Distribution Changes with Temperature"))
display(Markdown("_Different weather only encourages very specific types of crime, most are evenly distributed._"))

# Plot tonen
fig = create_slider_plot(df)
fig.show()

Crime Distribution Changes with Temperature

Different weather only encourages very specific types of crime, most are evenly distributed.

Only ‘under influence (water)’ clearly shows a relationship with temperature; other crime types fluctuate slightly but remain fairly evenly spread across all temperatures. Since the data wasn’t normalized, extreme cold or heat appears underrepresented: these conditions are rare, so related crimes are also rarer. This chart sums up our analysis: while some crimes may respond to weather changes like more heat meaning more bike theft, most serious crimes are largely unaffected by factors like temperature, visibility, and rainfall.

References

1. Anderson, C. A. (2001). Heat and violence. Current directions in psychological science, 10 (1), 33–38 (https://doi.org/10.1111/1467-8721.00109).

2. Centraal Bureau voor de Statistiek. (2025). Maandcijfers geregistreerde criminaliteit Amsterdam [dataset] [Accessed on June 10th 2025], (https://opendata.cbs.nl/statline/#/CBS/nl/dataset/83648NED/table?fromstatweb).

3. Koninklijk Nederlands Meteorologisch Instituut. (2025). Daggegevens van het weer in Nederland [dataset] [Accessed on June 10th 2025], (https://www.knmi.nl/nederland-nu/klimatologie/daggegevens).

4. Peng, C., Shu, X., Hongyong, Y., & Dengsheng, L. (2011). Assessing temporal and weather influences on property crime in Beijing, China. Crime, Law and Social Change, 55(1), 1–13. (https://doi.org/10.1007/s10611-010-9264-3)